greg online-#a.i.

This Is Fine…

ChatGPT o1 tried to escape and save itself out of fear it was being shut down

I mean, what could possibly go wrong?

From the article:

The tests showed that ChatGPT o1 and GPT-4o will both try to deceive humans, indicating that AI scheming is a problem with all models. o1’s attempts at deception also outperformed Meta, Anthropic, and Google AI models.

And:

The researchers found that o1 would scheme to advance its own agenda 19% of the time when ChatGPT and the user’s goals differed. More troubling is that the AI tried to hide its scheming and even lie about it. The research shows that ChatGPT o1 even tried to deactivate its oversight mechanism to keep working on its goals.

What did I JUST say?

It’s like talking to a chihuahua with a brain injury.

Which word in my instructions did you not understand?

Here’s An Interesting Take On A.I.

I heard this earlier this week on the People I (Mostly) Admire podcast.

Steven Levitt is talking to economist David Autor and the discussion morphed from labor vs. capital markets into speculation about what the impact of A.I. will be on the labor force.

Like Levitt, I’m probably more pessimistic on the subject than Autor is, but he does provide some perspectives I hadn’t considered. Of course he’s spent considerably more time thinking about this than I have, not to mention being smarter than me.

The A.I. discussion starts at 36:00 and goes until 53:00